On Fri, Jul 24, 2026 at 10:41:10PM +0300, Aigars Mahinovs wrote: > On Fri, 24 Jul 2026, 18:41 The Wanderer, <[email protected]> wrote: > > > Whether or not that's a realistic concern (then or now, since IIRC there > > may have been some rulings which could have gone that way but didn't?) > > is a different matter, but that at least seems like a distinctly > > different consideration from "can it be copyrighted at all?". > > AFAIK Bartz v. Anthropic is one such landmark case - > https://legalblogs.wolterskluwer.com/copyright-blog/the-bartz-v-anthropic-settlement-understanding-americas-largest-copyright-settlement/ > has some analysis and background, for example. > > The key finding, for our case was the ruling from last year: "that > Anthropic's use of legally acquired books for AI training was > "quintessentially transformative" and protected as fair use". > > This basically is a bedrock decision that established that copyright of the > training materials do NOT survive the transformation inherent in the > learning process and thus the output of an LLM is NOT a derivative work of > the training material in the legal sense. >...
A ruling by a single district judge is not a bedrock decision, a district judge in a different district[1] discusses ChatGPT output for works by George R. R. Martin whose copyright was infringed during training for stating: There is no doubt that a reasonable jury applying the more discerning observer test could determine that this output is substantially similar to Martin’s original work based on the output’s incorporation of such copyrightable elements of Martin’s original work as setting, plot, and characters.[2] In Kadrey v. Meta (same district as your case), the judge wrote that the plaintiffs might have won had they argued that the LLM output competes with their own work: As for the potentially winning argument — that Meta has copied their works to create a product that will likely flood the market with similar works, causing market dilution — the plaintiffs barely give this issue lip service, and they present no evidence about how the current or expected outputs from Meta’s models would dilute the market for their own works.[3] The situation in the US is far from settled, and this does not even touch how the legal situation in the rest of the world might be where the fair use defense used in the US trials does not exist. Permitting LLM output in Debian is problematic from a legal point of view, but being on the cautious side is not a workable option since for a lot of upstream software that would pin us to ~ trixie without security updates. > Best Regards, > > Aigars Mahinovs. cu Adrian [1] fun fact: in both cases the judges were born in 1945 [2] https://assets.law360news.com/2404000/2404371/https-ecf-nysd-uscourts-gov-doc1-127138452540.pdf [3] https://cases.justia.com/federal/district-courts/california/candce/3:2023cv03417/415175/598/0.pdf
